Sampling an Edge Uniformly in Sublinear Time

TL;DR

This paper presents an efficient algorithm for uniformly sampling edges in sublinear time without direct access to random edge queries, advancing the understanding of graph sampling models and enabling better simulation of sublinear algorithms.

Contribution

It provides the first algorithm to sample edges uniformly in expected sublinear time without random edge access, with improved bounds and a method to approximate uniform sampling.

Findings

Expected time $O(rac{n}{\sqrt{m}} \, ext{log} \, n)$ for uniform edge sampling.

Improved algorithm for $\epsilon$-close to uniform sampling in expected time $O(rac{n}{\sqrt{m}} \, ext{log} \, rac{1}{\epsilon})$.

Sampling from near-uniform distributions suffices to simulate sublinear algorithms with minimal success probability loss.

Abstract

The area of sublinear algorithms have recently received a lot of attention. In this setting, one has to choose specific access model for the input, as the algorithm does not have time to pre-process or even to see the whole input. A fundamental question remained open on the relationship between the two common models for graphs -- with and without access to the "random edge" query -- namely whether it is possible to sample an edge uniformly at random in the model without access to the random edge queries. In this paper, we answer this question positively. Specifically, we give an algorithm solving this problem that runs in expected time $O(\frac{n}{\sqrt{m}} \log n)$. This is only a logarithmic factor slower than the lower bound given in [5]. Our algorithm uses the algorithm from [7] which we analyze in a more careful way, leading to better bounds in general graphs. We also show a way to sample edges $\epsilon$-close to uniform in expected time $O(\frac{n}{\sqrt{m}} \log \frac{1}{\epsilon})$, improving upon the best previously known algorithm. We also note that sampling edges from a distribution sufficiently close to uniform is sufficient to be able to simulate sublinear algorithms that use the random edge queries while decreasing the success probability of the algorithm only by $o(1)$. This allows for a much simpler algorithm that can be used to emulate random edge queries.